Since it began in 2012, PREDICTS has furthered our understanding of how different aspects of biodiversity respond to land-use pressures across the globe. The first phase of PREDICTS focussed on collating spatial comparisons, where biodiversity was sampled across multiple sites that differed in land use or intensity. This approach had the advantage that there are many such comparisons in the literature and thanks to the generosity of hundreds of data contributors, we were able to develop a taxonomically and geographically representative database to underpin global models of how biodiversity responds to land use. However, there are limits to what we can learn about land-use impacts from data like these, because they can’t tell us directly how diversity has changed over time. We are therefore working a second phase of PREDICTS, funded by NERC, in which we are trying to collate biodiversity data from temporal comparisons, ideally where terrestrial sites have been surveyed both before and after a change in land use or land-use intensity. Adriana De Palma is the postdoc on this phase of the project; Samantha Hill and Sara Contu are informatics technicians and Andy Purvis is the principal investigator.
We also have a wide range of project partners from many different organisations: Amanda Bates, Luca Börger, Neil Burgess, Stuart Butchart, Simon Ferrier, Jens Kattge, Phil Martin, Tim Newbold, Samraat Pawar, Will Pearse and Jorn Scharlemann.
Our main questions are:
- Which happens first when land use changes: loss of current diversity or gain of new diversity?
- Which aspects of assemblage structure and function are most sensitive to land-use change?
- How much difference does ‘biotic lag’ make to the models from PREDICTS so far?
- When do species’ traits and phylogeny predict their responses to change?
- Which species are both sensitive and exposed to land-use change?
- What mediates the resistance or resilience of a community to a land-use change?
- How quickly do communities recover after a land-use pressure is removed?
As with “PREDICTS v1”, we aim to develop an extensive, openly available database of such biodiversity data and all contributors will again be offered co-authorship on a manuscript that describes that database. We would love to hear from any researchers who would like to contribute data to PREDICTS v2 (please email any offers to email@example.com). We are particularly interested in obtaining before-after control-impact studies, but are also looking for before-after comparisons lacking control sites, and control-impact studies that sample for several years after a land-use change (the timing of which is known). The database is growing steadily. At the last tally, we had 76 sources in the database, which relate to 84 studies, 3159 sites and 9172 species; in total, we have 1,489,278 rows of data, which span a range of land-use transitions including: agricultural abandonment and de-intensification, habitat restoration, logging activities and deforestation, and the establishment of biofuel crops.
Many thanks to all those who have contributed data so far and to all the students who have worked on the project. So far, seven students have successfully completed postgraduate projects as part of this new phase of PREDICTS.
A project of projections
The P in PREDICTS is for Projecting. That’s because it is no longer sufficient to model biodiversity; we must also be able to project the effects of human-driven changes on biodiversity. Unfortunately, our initial efforts to project biodiversity using PREDICTS-derived models, in space and time, were cumbersome and slow. When the team did a high-resolution projection (cell size of 30 arc-degree seconds) the effort took multiple weeks and caused one of the lab servers to crash! As part of his MRes project at Imperial College, Ricardo Gonzalez has developed a generic set of tools that simplify and accelerate generating projections based on linear or mixed-effects models developed in R. Using this system it is possible to do the same high-resolution projection in about 5 minutes (down from several weeks). Alternatively, projecting biodiversity for the rest of the century at a coarser spatial scale for any scenario takes a few minutes and generates a nice video for visualizing the projections. And when someone develops a new model, they can immediately generate projections without having to modify the code. The framework consists of three parts: a library for lazily evaluating functions on raster maps; a library for extracting R models from RDS files for use in Python or C; and a shim layer with PREDICTS-specific definitions. The first two are not tied to PREDICTS and can be used with any linear or mixed-effects model in R. Only that last part is PREDICTS-specific. The system generates the projection in a CPU and memory efficient way, which is especially important for high resolution projections. Ricardo is using this framework for projecting biodiversity (both abundance and species richness), for all Shared Socioeconomic Pathways (SSPs) scenarios for which there are harmonized land use data. The code will be pushed to GitHub after completing the project, but if you would like to use this framework before then, please get in touch.